Of the enormous craze on the art of AI to Human robots2022 witnessed a paradigm change in technology. Meta Ai, a pioneer in the field, made sure that he was not left to transform the industry.
Here are Meta AI’s innovations that helped him stay with the white AI epicenter in 2022!
What awaits us in AI generating art?
An image is worth a thousand words. But is it? With the text generation in the image, a few words may be sufficient to create a thousand images.
Meta challenged the monotony of the text generation process in the image with its own AI model, ‘Make-a-Scene’It not only takes text prompts but also sketches to create high-definition visual masterpieces on a digital canvas. In addition, later this year, Meta IA researchers progressed to generate art by invites by announcing Make-A-VideoA new AI system that transforms the text into a brief and silent video clips.
And that’s not it.
Isn’t the name Data2 with Does that seem familiar? It is probably because he splashed the media in January. This is an important entry into the growing list of strategies focused on the creation of unique automatic learning architectures that manage many different types of data, such as text, image and speech.
“The first high performance self-supervised algorithm that works for several methods”, as Meta called it. The researchers applied him separately to speech, text and images, where he surpassed the best single -use algorithms for computer vision and speech. He became competitive on NLP tasks.
Meta has added that Data2 with is not based on contrastive learning or reconstruction of the input example. The technology giant has also published open source code and pre-trained models.
How is the 50-year problem of protein structures solved?
When Depth Predicted structures revealed for some 220 million proteins, the trove has covered almost all the proteins of organisms known in DNA databases. By taking the research in advance, Meta Ai fills the ‘dark matter of the protein universe‘.
(Source: Meta))
Meta researchers used AI to predict the structures of some 600 million proteins from bacteria, viruses and other microorganisms that have not been characterized.
The team created the first database revealing the structures of the metagenomic world on the enormous scale of hundreds of millions of proteins. The predictions were made using a “large language model”, a basic tool to predict the text from a few words.
How can AI improve the life cycle of automatic learning?
To use AI more efficiently, several challenges must be met by the community. The system must be able to adapt to engineers without ML backgrounds, and it must provide mechanisms to optimize a variety of product objectives.
Meta Researchers have developedLoop“An end-to-end platform with easy-to-use APIs for optimization, personalization and comment collection to meet these needs. Looper can be used to support the entire life cycle of automatic learning, from model training to deployment and inference to the evaluation and optimization of products. In addition, Looper allows us to modify existing products to take advantage of AI for personalized optimizations rather than having to rebuild them around AI models.
(Source: Meta))
Currently, the platform hosts 700 models AI and produces 4 million AI outputs every second.
How can AI improve human communication?
IA translation has mainly focused on written languages. However, about half of the 7,000 languages of the world are mainly oral – without standard or widely used writing system. Thus, it is impossible to create automatic translation tools using standard techniques requiring large quantities of written text to form an AI model.
To take up this challenge, Meta built the First translation system powered by AI For mainly oral language – Hokkien – widely spoken in the Chinese diaspora. Meta technology allows Hokkien speakers to converse with English speakers because the language lacks a standard written form.
In addition, Meta released her No languages left behind Project for building an AI model which can translate more and more languages, even those which are less commonly spoken. Announced for the first time in March, the AI model can now translate into 200 languages, said the technology giant.
Appealing to the “superpower of AI,” said meta-PDG Mark Zuckerberg In a Facebook publication that it will be used in the various Meta products and services and will also be available on Github So that anyone can use.